AI agents, automations, and chatbots, explained in plain English

AI vocabulary has become unusually flexible. A chatbot gets called an agent. An automation gets described as an AI employee. A familiar search box becomes a copilot. The labels make simple products sound ambitious, but they do not tell you what the system can actually access, decide, or do.

You do not need to police every term. You do need enough of a working definition to understand what you are buying and how much responsibility the system is being given.

A chatbot holds a conversation

A chatbot is an interface. You type or speak, and it responds in the same conversation. A basic support bot might answer from a list of approved questions. A generative chatbot can interpret less predictable wording and compose a new response from its instructions and available context.

The conversation alone does not make it an agent. A chatbot may only answer questions. It may also sit in front of an automation or an agent that can search company documents, create a task, or update a record. The chat window tells you how a person interacts with the system, not how the system works behind it.

An automation follows a defined path

An automation performs a sequence that people designed in advance. A form submission creates a contact in the CRM, assigns an owner based on region, and sends a notification. The steps and conditions are known before the workflow runs.

Conventional automations are excellent when the rules are stable. They are fast, predictable, and relatively easy to test. They can also include an AI step. A model might summarize a long form response or classify its topic before the workflow continues along a fixed path. Adding that step does not necessarily turn the whole system into an agent.

An agent chooses how to pursue a goal

An AI agent has tools it can use and some freedom to decide what to do next. Instead of following one fixed sequence, it can plan a step, take an action, inspect the result, and adjust its approach. It may search several sources, notice missing information, ask a follow-up question, or choose a different tool when the first attempt does not work.

That flexibility is useful when the route to a good outcome cannot be fully prescribed. It also introduces more uncertainty. Each tool, decision, and retry creates another place where the system can misunderstand the goal, use the wrong information, or take an action a person did not expect.

The same product can contain all three

Imagine a marketing team asking for help with a campaign brief. The person opens a chatbot and describes the campaign. A fixed automation collects the approved product facts and audience details. An agent searches research sources, identifies gaps, and asks for the missing context. Another automation saves the draft in the project system and routes it to a human reviewer.

Calling the entire product an agent is not necessarily wrong, but the label hides the useful design questions. Which steps are fixed? Which choices belong to the model? What can it read? What can it change? When does a person approve the result?

More autonomy is not automatically better

A system that can make more decisions may handle more variation, but it is also harder to predict, test, and govern. It may take longer to run and cost more because the model uses several steps instead of one. For a stable process with clear rules, an agent can be an expensive way to recreate a dependable automation.

Use autonomy where it solves a real problem. Research across changing sources, investigating an unfamiliar issue, or coordinating a task with many possible paths may justify an agent. Moving approved data between two systems probably does not.

Questions that make the label less important

• What information can the system read, and does that access follow the user's permissions?
• Can it change records, publish content, send messages, spend money, or only prepare a draft?
• Which steps always follow fixed rules, and which decisions does the model make?
• What requires approval before the system continues?
• How are its sources, decisions, tool use, and failures recorded?
• What happens when information is missing or the system is uncertain?
• Can a person stop, correct, and safely retry the work?

Choose the simplest useful shape

If the path is predictable, begin with an automation. If people need a convenient way to ask questions, a chatbot may be enough. If the work has a variable path and benefits from choosing among tools, consider an agent with clear limits and review points.

The best system may combine them, but it should not become more complicated simply to earn a more fashionable name. Start with the job, give the technology only the access and freedom it needs, and make sure someone remains accountable for the result.